Official Pytorch implementation of HappyQuokka system that submitted to the 2023 ICASSP Auditory EEG challenge task 2: regression.
This repository is based on FastSpeech github (Paper).
python train.py --experiment_foler {Your experiment name}
In train.py, change --dataset_folder to the absolute path of the dataset directory.
- Auxiliary global conditioner only used for within-subjects generation.
- When generating stimulus for heldout-subjects, please change
--g_conintrain.pyinto False. - For a quick start, you can refer to Auditory EEG challenge github and EEG dataset, download the split_data.zip for experiment.
@article{fastspeech,
title={Fastspeech: Fast, robust and controllable text to speech},
author={Ren, Yi and Ruan, Yangjun and Tan, Xu and Qin, Tao and Zhao, Sheng and Zhao, Zhou and Liu, Tie-Yan},
journal={Advances in neural information processing systems},
volume={32},
year={2019}
}
@inproceedings{prelayernorm,
title={On layer normalization in the transformer architecture},
author={Xiong, Ruibin and Yang, Yunchang and He, Di and Zheng, Kai and Zheng, Shuxin and Xing, Chen and Zhang, Huishuai and Lan, Yanyan and Wang, Liwei and Liu, Tieyan},
booktitle={International Conference on Machine Learning},
pages={10524--10533},
year={2020},
organization={PMLR}
}
@data{eegdata_K3VSND_2023,
author = {Bollens, Lies and Accou, Bernd and Van hamme, Hugo and Francart, Tom},
publisher = {KU Leuven RDR},
title = {{A Large Auditory EEG decoding dataset}},
year = {2023},
version = {V1},
doi = {10.48804/K3VSND},
url = {https://doi.org/10.48804/K3VSND}
}